An integrated model of acting and inference

Abstract

This paper discusses propositional representations for representing beliefs, actions, and plans. From the point of integrating inference and acting, it introduces the notions of a transformer. A transformer is a representation that specifies a belief/act transformation under the influence of a transforming procedure. We will look at forward and backward transforming procedures. Under this scheme a reasoning rule in SNePS is a belief-belief transformer, i.e. it is a specification of antecedent and consequent beliefs. Such a transformer can be used in forward as well as backward inference (using the transformation procedures). Similarly, we can have belief-action, action-belief, and action-action transformers. These transformers can be used to model planning, plan decompositions, sequencing of actions, effects and preconditions of actions, and reactivity to sensory input. While some of these transformers can be uniformly used by both forward as well as backward transformation procedures, it is not desirable to do so. For this reason we need different representations for same transformers so that some transformations can be blocked. We informally outline a scheme for modifying the existing inference procedure (SNIP) to do forward/backward transformations involved in reasoning as well as acting.